Source code for openquake.hazardlib.gsim.chiou_youngs_2014

# -*- coding: utf-8 -*-
# vim: tabstop=4 shiftwidth=4 softtabstop=4
#
# Copyright (C) 2012-2023 GEM Foundation
#
# OpenQuake is free software: you can redistribute it and/or modify it
# under the terms of the GNU Affero General Public License as published
# by the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# OpenQuake is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with OpenQuake. If not, see <http://www.gnu.org/licenses/>.

"""
Module exports :class:`ChiouYoungs2014`
               :class:`ChiouYoungs2014Japan`
               :class:`ChiouYoungs2014Italy`
               :class:`ChiouYoungs2014Wenchuan`
               :class:`ChiouYoungs2014PEER`
               :class:`ChiouYoungs2014NearFaultEffect`
"""
import numpy as np

from openquake.baselib.general import CallableDict
from openquake.hazardlib.gsim.base import GMPE, CoeffsTable
from openquake.hazardlib.gsim.abrahamson_2014 import get_epistemic_sigma
from openquake.hazardlib import const
from openquake.hazardlib.imt import PGA, PGV, SA

CONSTANTS = {"c2": 1.06, "c4": -2.1, "c4a": -0.5, "crb": 50.0,
             "c8a": 0.2695, "c11": 0.0, "phi6": 300.0, "phi6jp": 800.0}


def _get_centered_cdpp(clsname, ctx):
    """
    Returns the centred dpp term (zero by default)
    """
    if clsname.endswith("NearFaultEffect"):
        return ctx.rcdpp
    return np.zeros(ctx.rrup.shape)


def _get_centered_z1pt0(clsname, ctx):
    """
    Get z1pt0 centered on the Vs30- dependent average z1pt0(m)
    California and non-Japan regions
    """
    if clsname.endswith("Japan"):
        mean_z1pt0 = (-5.23 / 2.) * np.log(((ctx.vs30 ** 2.) + 412.39 ** 2.)
                                           / (1360 ** 2. + 412.39 ** 2.))
        return ctx.z1pt0 - np.exp(mean_z1pt0)

    #: California and non-Japan regions
    mean_z1pt0 = (-7.15 / 4.) * np.log(((ctx.vs30) ** 4. + 570.94 ** 4.)
                                       / (1360 ** 4. + 570.94 ** 4.))
    return ctx.z1pt0 - np.exp(mean_z1pt0)


def _get_centered_ztor(ctx):
    """
    Get ztor centered on the M- dependent avarage ztor(km)
    by different fault types.
    """
    # Strike-slip and normal faulting
    mean_ztor = np.clip(2.673 - 1.136 * np.clip(ctx.mag - 4.970, 0., None),
                        0., None) ** 2
    # Reverse and reverse-oblique faulting
    rev = (30. <= ctx.rake) & (ctx.rake <= 150.)
    mean_ztor[rev] = np.clip(2.704 - 1.226 * np.clip(
        ctx.mag[rev] - 5.849, 0.0, None), 0., None) ** 2
    return ctx.ztor - mean_ztor


def _get_ln_y_ref(ctx, C):
    """
    Get an intensity on a reference soil.
    Implements eq. 13a.
    """
    # Reverse faulting flag
    Frv = 1. if 30 <= ctx.rake <= 150 else 0.
    # Normal faulting flag
    Fnm = 1. if -120 <= ctx.rake <= -60 else 0.
    # A part in eq. 11
    mag_test1 = np.cosh(2. * np.clip(ctx.mag - 4.5, 0., None))
    # Centered DPP
    centered_dpp = 0
    # Centered Ztor
    centered_ztor = 0

    dist_taper = np.fmax(1 - (np.fmax(ctx.rrup - 40,
                              np.zeros_like(ctx)) / 30.),
                         np.zeros_like(ctx))
    dist_taper = dist_taper.astype(np.float64)
    ln_y_ref = (
        # first part of eq. 11
        C['c1']
        + (C['c1a'] + C['c1c'] / mag_test1) * Frv
        + (C['c1b'] + C['c1d'] / mag_test1) * Fnm
        + (C['c7'] + C['c7b'] / mag_test1) * centered_ztor
        + (C['c11'] + C['c11b'] / mag_test1) *
        np.cos(np.radians(ctx.dip)) ** 2
        # second part
        + C['c2'] * (ctx.mag - 6)
        + ((C['c2'] - C['c3']) / C['cn'])
        * np.log(1 + np.exp(C['cn'] * (C['cm'] - ctx.mag)))
        # third part
        + C['c4']
        * np.log(ctx.rrup + C['c5']
                 * np.cosh(C['c6'] * np.clip(ctx.mag - C['chm'], 0, None)))
        + (C['c4a'] - C['c4'])
        * np.log(np.sqrt(ctx.rrup ** 2 + C['crb'] ** 2))
        # forth part
        + (C['cg1'] + C['cg2'] / (
            np.cosh(np.clip(ctx.mag - C['cg3'], 0, None))))
        * ctx.rrup
        # fifth part
        + C['c8'] * dist_taper
        * np.clip((ctx.mag - 5.5, 0) / 0.8, 0., 1.)
        * np.exp(-1 * C['c8a'] * (ctx.mag - C['c8b']) ** 2) * centered_dpp
        # sixth part
        # + C['c9'] * Fhw * np.cos(math.radians(ctx.dip)) *
        # (C['c9a'] + (1 - C['c9a']) * np.tanh(ctx.rx / C['c9b']))
        # * (1 - np.sqrt(ctx.rjb ** 2 + ctx.ztor ** 2)
        #   / (ctx.rrup + 1.0))
    )
    return ln_y_ref


def _get_mean(ctx, C, ln_y_ref, exp1, exp2):
    """
    Add site effects to an intensity. Implements eq. 13b.
    """
    eta = epsilon = 0.
    ln_y = (
        # first line of eq. 12
        ln_y_ref + eta
        # second line
        + C['phi1'] * np.log(ctx.vs30 / 1130).clip(-np.inf, 0)
        # third line
        + C['phi2'] * (exp1 - exp2)
        * np.log((np.exp(ln_y_ref) * np.exp(eta) + C['phi4']) / C['phi4'])
        # fourth line - removed
        # fifth line
        + epsilon)
    return ln_y


[docs]def get_basin_depth_term(clsname, C, centered_z1pt0): """ Returns the basin depth scaling """ if clsname.endswith("Japan"): return C["phi5jp"] * (1.0 - np.exp(-centered_z1pt0 / CONSTANTS["phi6jp"])) return C["phi5"] * (1.0 - np.exp(-centered_z1pt0 / CONSTANTS["phi6"]))
[docs]def get_directivity(clsname, C, ctx): """ Returns the directivity term. The directivity prediction parameter is centered on the average directivity prediction parameter. Here we set the centered_dpp equal to zero, since the near fault directivity effect prediction is off by default in our calculation. """ cdpp = _get_centered_cdpp(clsname, ctx) if not np.any(cdpp > 0.0): # No directivity term return 0.0 f_dir = np.exp(-C["c8a"] * ((ctx.mag - C["c8b"]) ** 2.)) * cdpp f_dir *= np.clip((ctx.mag - 5.5) / 0.8, 0., 1.) rrup_max = ctx.rrup - 40. rrup_max[rrup_max < 0.0] = 0.0 rrup_max = 1.0 - (rrup_max / 30.) rrup_max[rrup_max < 0.0] = 0.0 return C["c8"] * rrup_max * f_dir
get_far_field_distance_scaling = CallableDict()
[docs]@get_far_field_distance_scaling.add("CAL") def get_far_field_distance_scaling_1(region, C, mag, rrup): """ Returns the far-field distance scaling term - both magnitude and distance - for California and other regions """ # Get the attenuation distance scaling f_r = (CONSTANTS["c4a"] - CONSTANTS["c4"]) * np.log( np.sqrt(rrup ** 2. + CONSTANTS["crb"] ** 2.)) # Get the magnitude dependent term f_rm = C["cg1"] + C["cg2"] / np.cosh(np.clip(mag - C["cg3"], 0.0, None)) return f_r + f_rm * rrup
[docs]@get_far_field_distance_scaling.add("JPN") def get_far_field_distance_scaling_2(region, C, mag, rrup): """ Returns the far-field distance scaling term - both magnitude and distance - for Japan """ # Get the attenuation distance scaling f_r = (CONSTANTS["c4a"] - CONSTANTS["c4"]) * np.log( np.sqrt(rrup ** 2. + CONSTANTS["crb"] ** 2.)) # Get the magnitude dependent term f_rm = (C["cg1"] + C["cg2"] / np.cosh(np.clip(mag - C["cg3"], 0.0, None))) * rrup # Apply adjustment factor for Japan f_rm[(mag > 6.0) & (mag < 6.9)] *= C["gjpit"] return f_r + f_rm
[docs]@get_far_field_distance_scaling.add("ITA") def get_far_field_distance_scaling_3(region, C, mag, rrup): """ Returns the far-field distance scaling term - both magnitude and distance - for Italy """ # Get the attenuation distance scaling f_r = (CONSTANTS["c4a"] - CONSTANTS["c4"]) * np.log( np.sqrt(rrup ** 2. + CONSTANTS["crb"] ** 2.)) # Get the magnitude dependent term f_rm = (C["cg1"] + C["cg2"] / np.cosh(np.clip(mag - C["cg3"], 0.0, None))) * rrup # Apply adjustment factor for Italy f_rm[(mag > 6.0) & (mag < 6.9)] *= C["gjpit"] return f_r + f_rm
[docs]@get_far_field_distance_scaling.add("WEN") def get_far_field_distance_scaling_4(region, C, mag, rrup): """ Returns the far-field distance scaling term - both magnitude and distance - for Wenchuan """ # Get the attenuation distance scaling f_r = (CONSTANTS["c4a"] - CONSTANTS["c4"]) * np.log( np.sqrt(rrup ** 2. + CONSTANTS["crb"] ** 2.)) # Get the magnitude dependent term f_rm = (C["cg1"] + C["cg2"] / np.cosh(np.clip(mag - C["cg3"], 0.0, None))) * rrup # Apply adjustment factor for Wenchuan return f_r + (f_rm * C["gwn"])
[docs]def get_geometric_spreading(C, mag, rrup): """ Returns the near-field geometric spreading term """ # Get the near-field magnitude scaling return CONSTANTS["c4"] * np.log( rrup + C["c5"] * np.cosh(C["c6"] * np.clip(mag - C["chm"], 0.0, None)))
[docs]def get_hanging_wall_term(C, ctx): """ Returns the hanging wall term """ fhw = np.zeros(ctx.rrup.shape) idx = ctx.rx >= 0.0 if np.any(idx): fdist = 1.0 - (np.sqrt(ctx.rjb[idx] ** 2. + ctx.ztor[idx] ** 2.) / (ctx.rrup[idx] + 1.0)) fdist *= C["c9a"] + (1.0 - C["c9a"]) * np.tanh(ctx.rx[idx] / C["c9b"]) fhw[idx] += C["c9"] * np.cos(np.radians(ctx.dip[idx])) * fdist return fhw
[docs]def get_linear_site_term(clsname, C, ctx): """ Returns the linear site scaling term """ if clsname.endswith("Japan"): return C["phi1jp"] * np.log(ctx.vs30 / 1130).clip(-np.inf, 0.0) return C["phi1"] * np.log(ctx.vs30 / 1130).clip(-np.inf, 0.0)
[docs]def get_region(clsname): if clsname.endswith("Italy"): return "ITA" elif clsname.endswith("Japan"): return "JPN" elif clsname.endswith("Wenchuan"): return "WEN" else: return "CAL"
[docs]def get_ln_y_ref(clsname, C, ctx): """ Returns the ground motion on the reference rock, described fully by Equation 11 """ region = get_region(clsname) delta_ztor = _get_centered_ztor(ctx) return (get_stress_scaling(C) + get_magnitude_scaling(C, ctx.mag) + get_source_scaling_terms(C, ctx, delta_ztor) + get_hanging_wall_term(C, ctx) + get_geometric_spreading(C, ctx.mag, ctx.rrup) + get_far_field_distance_scaling(region, C, ctx.mag, ctx.rrup) + get_directivity(clsname, C, ctx))
[docs]def get_magnitude_scaling(C, mag): """ Returns the magnitude scaling """ f_m = np.log(1.0 + np.exp(C["cn"] * (C["cm"] - mag))) f_m = CONSTANTS["c2"] * (mag - 6.0) +\ ((CONSTANTS["c2"] - C["c3"]) / C["cn"]) * f_m return f_m
[docs]def get_nonlinear_site_term(C, ctx, y_ref): """ Returns the nonlinear site term and the Vs-scaling factor (to be used in the standard deviation model """ vs = ctx.vs30.clip(-np.inf, 1130.0) f_nl_scaling = C["phi2"] * (np.exp(C["phi3"] * (vs - 360.)) - np.exp(C["phi3"] * (1130. - 360.))) f_nl = np.log((y_ref + C["phi4"]) / C["phi4"]) * f_nl_scaling return f_nl, f_nl_scaling
[docs]def get_phi(C, mag, ctx, nl0): """ Returns the within-event variability described in equation 13, line 3 """ phi = C["sig3"] * np.ones(ctx.vs30.shape) phi[ctx.vs30measured] = 0.7 phi = np.sqrt(phi + ((1.0 + nl0) ** 2.)) mdep = C["sig1"] + ( C["sig2"] - C["sig1"]) * np.clip(mag - 5., 0., 1.5) / 1.5 return mdep * phi
[docs]def get_source_scaling_terms(C, ctx, delta_ztor): """ Returns additional source scaling parameters related to style of faulting, dip and top of rupture depth """ f_src = np.zeros_like(ctx.mag) coshm = np.cosh(2.0 * np.clip(ctx.mag - 4.5, 0., None)) # Style of faulting term pos = (30 <= ctx.rake) & (ctx.rake <= 150) neg = (-120 <= ctx.rake) & (ctx.rake <= -60) # reverse faulting flag f_src[pos] += C["c1a"] + (C["c1c"] / coshm[pos]) # normal faulting flag f_src[neg] += C["c1b"] + (C["c1d"] / coshm[neg]) # Top of rupture term f_src += (C["c7"] + (C["c7b"] / coshm)) * delta_ztor # Dip term f_src += ((CONSTANTS["c11"] + (C["c11b"] / coshm)) * np.cos(np.radians(ctx.dip)) ** 2.0) return f_src
[docs]def get_stddevs(clsname, C, ctx, mag, y_ref, f_nl_scaling): """ Returns the standard deviation model described in equation 13 """ if clsname == 'ChiouYoungs2014PEER': # the standard deviation, which is fixed at 0.65 for every site return [0.65 * np.ones_like(ctx.vs30), 0, 0] # Determines the nonlinear term described in equation 13, line 4 nl0 = f_nl_scaling * (y_ref / (y_ref + C["phi4"])) # Get between and within-event variability tau = get_tau(C, mag) phi_nl0 = get_phi(C, mag, ctx, nl0) # Get total standard deviation propagating the uncertainty in the # nonlinear amplification term sigma = np.sqrt(((1.0 + nl0) ** 2.) * (tau ** 2.) + phi_nl0 ** 2.) return [sigma, np.abs((1 + nl0) * tau), phi_nl0]
[docs]def get_stress_scaling(C): """ Returns the stress drop scaling factor """ return C["c1"]
[docs]def get_tau(C, mag): """ Returns the between-event variability described in equation 13, line 2 """ # eq. 13 to calculate inter-event standard error mag_test = np.clip(mag - 5.0, 0., 1.5) return C['tau1'] + (C['tau2'] - C['tau1']) / 1.5 * mag_test
[docs]def get_mean_stddevs(name, C, ctx): """ Return mean and standard deviation values """ # Get ground motion on reference rock ln_y_ref = get_ln_y_ref(name, C, ctx) y_ref = np.exp(ln_y_ref) # Get the site amplification # Get basin depth dz1pt0 = _get_centered_z1pt0(name, ctx) # for Z1.0 = 0.0 no deep soil correction is applied dz1pt0[ctx.z1pt0 <= 0.0] = 0.0 f_z1pt0 = get_basin_depth_term(name, C, dz1pt0) # Get linear amplification term f_lin = get_linear_site_term(name, C, ctx) # Get nonlinear amplification term f_nl, f_nl_scaling = get_nonlinear_site_term(C, ctx, y_ref) # Add on the site amplification mean = ln_y_ref + (f_lin + f_nl + f_z1pt0) # Get standard deviations sig, tau, phi = get_stddevs( name, C, ctx, ctx.mag, y_ref, f_nl_scaling) return mean, sig, tau, phi
[docs]class ChiouYoungs2014(GMPE): """ Implements GMPE developed by Brian S.-J. Chiou and Robert R. Youngs Chiou, B. S.-J. and Youngs, R. R. (2014), "Updated of the Chiou and Youngs NGA Model for the Average Horizontal Component of Peak Ground Motion and Response Spectra, Earthquake Spectra, 30(3), 1117 - 1153, DOI: 10.1193/072813EQS219M """ adapted = False # overridden in acme_2019 #: Supported tectonic region type is active shallow crust DEFINED_FOR_TECTONIC_REGION_TYPE = const.TRT.ACTIVE_SHALLOW_CRUST #: Supported intensity measure types are spectral acceleration, #: peak ground velocity and peak ground acceleration DEFINED_FOR_INTENSITY_MEASURE_TYPES = {PGA, PGV, SA} #: Supported intensity measure component is orientation-independent #: measure :attr:`~openquake.hazardlib.const.IMC.RotD50`, DEFINED_FOR_INTENSITY_MEASURE_COMPONENT = const.IMC.RotD50 #: Supported standard deviation types are inter-event, intra-event #: and total, see chapter "Variance model". DEFINED_FOR_STANDARD_DEVIATION_TYPES = { const.StdDev.TOTAL, const.StdDev.INTER_EVENT, const.StdDev.INTRA_EVENT} #: Required site parameters are Vs30, Vs30 measured flag #: and Z1.0. REQUIRES_SITES_PARAMETERS = {'vs30', 'vs30measured', 'z1pt0'} #: Required rupture parameters are magnitude, rake, #: dip and ztor. REQUIRES_RUPTURE_PARAMETERS = {'dip', 'rake', 'mag', 'ztor'} #: Required distance measures are RRup, Rjb and Rx. REQUIRES_DISTANCES = {'rrup', 'rjb', 'rx'} #: Reference shear wave velocity DEFINED_FOR_REFERENCE_VELOCITY = 1130 def __init__(self, sigma_mu_epsilon=0.0, **kwargs): super().__init__(sigma_mu_epsilon=sigma_mu_epsilon, **kwargs) self.sigma_mu_epsilon = sigma_mu_epsilon
[docs] def compute(self, ctx: np.recarray, imts, mean, sig, tau, phi): """ See :meth:`superclass method <.base.GroundShakingIntensityModel.compute>` for spec of input and result values. """ name = self.__class__.__name__ # reference to page 1144, PSA might need PGA value pga_mean, pga_sig, pga_tau, pga_phi = get_mean_stddevs(name, self.COEFFS[PGA()], ctx) for m, imt in enumerate(imts): if repr(imt) == "PGA": mean[m] = pga_mean mean[m] += (self.sigma_mu_epsilon*get_epistemic_sigma(ctx)) sig[m], tau[m], phi[m] = pga_sig, pga_tau, pga_phi else: imt_mean, imt_sig, imt_tau, imt_phi = \ get_mean_stddevs(name, self.COEFFS[imt], ctx) # reference to page 1144 # Predicted PSA value at T ≤ 0.3s should be set equal to the value of PGA # when it falls below the predicted PGA mean[m] = np.where(imt_mean < pga_mean, pga_mean, imt_mean) \ if repr(imt).startswith("SA") and imt.period <= 0.3 \ else imt_mean mean[m] += (self.sigma_mu_epsilon*get_epistemic_sigma(ctx)) sig[m], tau[m], phi[m] = imt_sig, imt_tau, imt_phi
#: Coefficient tables are constructed from values in tables 1 - 5 COEFFS = CoeffsTable(sa_damping=5, table="""\ IMT c1 c1a c1b c1c c1d cn cm c2 c3 c4 c4a crb c5 chm c6 c7 c7b c8 c8a c8b c9 c9a c9b c11 c11b cg1 cg2 cg3 phi1 phi2 phi3 phi4 phi5 phi6 gjpit gwn phi1jp phi5jp phi6jp tau1 tau2 sig1 sig2 sig3 sig2jp pga -1.5065 0.165 -0.255 -0.165 0.255 16.0875 4.9993 1.06 1.9636 -2.1 -0.5 50 6.4551 3.0956 0.4908 0.0352 0.0462 0. 0.2695 0.4833 0.9228 0.1202 6.8607 0. -0.4536 -0.007146 -0.006758 4.2542 -0.521 -0.1417 -0.00701 0.102151 0. 300 1.5817 0.7594 -0.6846 0.459 800. 0.4 0.26 0.4912 0.3762 0.8 0.4528 pgv 2.3549 0.165 -0.0626 -0.165 0.0626 3.3024 5.423 1.06 2.3152 -2.1 -0.5 50 5.8096 3.0514 0.4407 0.0324 0.0097 0.2154 0.2695 5. 0.3079 0.1 6.5 0 -0.3834 -0.001852 -0.007403 4.3439 -0.7936 -0.0699 -0.008444 5.41 0.0202 300. 2.2306 0.335 -0.7966 0.9488 800. 0.3894 0.2578 0.4785 0.3629 0.7504 0.3918 0.01 -1.5065 0.165 -0.255 -0.165 0.255 16.0875 4.9993 1.06 1.9636 -2.1 -0.5 50 6.4551 3.0956 0.4908 0.0352 0.0462 0. 0.2695 0.4833 0.9228 0.1202 6.8607 0. -0.4536 -0.007146 -0.006758 4.2542 -0.521 -0.1417 -0.00701 0.102151 0. 300 1.5817 0.7594 -0.6846 0.459 800. 0.4 0.26 0.4912 0.3762 0.8 0.4528 0.02 -1.4798 0.165 -0.255 -0.165 0.255 15.7118 4.9993 1.06 1.9636 -2.1 -0.5 50 6.4551 3.0963 0.4925 0.0352 0.0472 0. 0.2695 1.2144 0.9296 0.1217 6.8697 0. -0.4536 -0.007249 -0.006758 4.2386 -0.5055 -0.1364 -0.007279 0.10836 0. 300 1.574 0.7606 -0.6681 0.458 800. 0.4026 0.2637 0.4904 0.3762 0.8 0.4551 0.03 -1.2972 0.165 -0.255 -0.165 0.255 15.8819 4.9993 1.06 1.9636 -2.1 -0.5 50 6.4551 3.0974 0.4992 0.0352 0.0533 0. 0.2695 1.6421 0.9396 0.1194 6.9113 0. -0.4536 -0.007869 -0.006758 4.2519 -0.4368 -0.1403 -0.007354 0.119888 0. 300 1.5544 0.7642 -0.6314 0.462 800. 0.4063 0.2689 0.4988 0.3849 0.8 0.4571 0.04 -1.1007 0.165 -0.255 -0.165 0.255 16.4556 4.9993 1.06 1.9636 -2.1 -0.5 50 6.4551 3.0988 0.5037 0.0352 0.0596 0. 0.2695 1.9456 0.9661 0.1166 7.0271 0. -0.4536 -0.008316 -0.006758 4.296 -0.3752 -0.1591 -0.006977 0.133641 0. 300 1.5502 0.7676 -0.5855 0.453 800. 0.4095 0.2736 0.5049 0.391 0.8 0.4642 0.05 -0.9292 0.165 -0.255 -0.165 0.255 17.6453 4.9993 1.06 1.9636 -2.1 -0.5 50 6.4551 3.1011 0.5048 0.0352 0.0639 0. 0.2695 2.181 0.9794 0.1176 7.0959 0. -0.4536 -0.008743 -0.006758 4.3578 -0.3469 -0.1862 -0.006467 0.148927 0. 300 1.5391 0.7739 -0.5457 0.436 800. 0.4124 0.2777 0.5096 0.3957 0.8 0.4716 0.075 -0.658 0.165 -0.254 -0.165 0.254 20.1772 5.0031 1.06 1.9636 -2.1 -0.5 50 6.4551 3.1094 0.5048 0.0352 0.063 0. 0.2695 2.6087 1.026 0.1171 7.3298 0. -0.4536 -0.009537 -0.00619 4.5455 -0.3747 -0.2538 -0.005734 0.190596 0. 300 1.4804 0.7956 -0.4685 0.383 800. 0.4179 0.2855 0.5179 0.4043 0.8 0.5022 0.1 -0.5613 0.165 -0.253 -0.165 0.253 19.9992 5.0172 1.06 1.9636 -2.1 -0.5 50 6.8305 3.2381 0.5048 0.0352 0.0532 0. 0.2695 2.9122 1.0177 0.1146 7.2588 0. -0.4536 -0.00983 -0.005332 4.7603 -0.444 -0.2943 -0.005604 0.230662 0. 300 1.4094 0.7932 -0.4985 0.375 800. 0.4219 0.2913 0.5236 0.4104 0.8 0.523 0.12 -0.5342 0.165 -0.252 -0.165 0.252 18.7106 5.0315 1.06 1.9795 -2.1 -0.5 50 7.1333 3.3407 0.5048 0.0352 0.0452 0. 0.2695 3.1045 1.0008 0.1128 7.2372 0. -0.4536 -0.009913 -0.004732 4.8963 -0.4895 -0.3077 -0.005696 0.253169 0. 300 1.3682 0.7768 -0.5603 0.377 800. 0.4244 0.2949 0.527 0.4143 0.8 0.5278 0.15 -0.5462 0.165 -0.25 -0.165 0.25 16.6246 5.0547 1.06 2.0362 -2.1 -0.5 50 7.3621 3.43 0.5045 0.0352 0.0345 0. 0.2695 3.3399 0.9801 0.1106 7.2109 0. -0.4536 -0.009896 -0.003806 5.0644 -0.5477 -0.3113 -0.005845 0.266468 0. 300 1.3241 0.7437 -0.6451 0.379 800. 0.4275 0.2993 0.5308 0.4191 0.8 0.5304 0.17 -0.5858 0.165 -0.248 -0.165 0.248 15.3709 5.0704 1.06 2.0823 -2.1 -0.5 50 7.4365 3.4688 0.5036 0.0352 0.0283 0. 0.2695 3.4719 0.9652 0.115 7.2491 0. -0.4536 -0.009787 -0.00328 5.1371 -0.5922 -0.3062 -0.005959 0.26506 0. 300 1.3071 0.7219 -0.6981 0.38 800. 0.4292 0.3017 0.5328 0.4217 0.8 0.531 0.2 -0.6798 0.165 -0.2449 -0.165 0.2449 13.7012 5.0939 1.06 2.1521 -2.1 -0.5 50 7.4972 3.5146 0.5016 0.0352 0.0202 0. 0.2695 3.6434 0.9459 0.1208 7.2988 0. -0.444 -0.009505 -0.00269 5.188 -0.6693 -0.2927 -0.006141 0.255253 0. 300 1.2931 0.6922 -0.7653 0.384 800. 0.4313 0.3047 0.5351 0.4252 0.8 0.5312 0.25 -0.8663 0.165 -0.2382 -0.165 0.2382 11.2667 5.1315 1.06 2.2574 -2.1 -0.5 50 7.5416 3.5746 0.4971 0.0352 0.009 0. 0.2695 3.8787 0.9196 0.1208 7.3691 0. -0.3539 -0.008918 -0.002128 5.2164 -0.7766 -0.2662 -0.006439 0.231541 0. 300 1.315 0.6579 -0.8469 0.393 800. 0.4341 0.3087 0.5377 0.4299 0.7999 0.5309 0.3 -1.0514 0.165 -0.2313 -0.165 0.2313 9.1908 5.167 1.06 2.344 -2.1 -0.5 50 7.56 3.6232 0.4919 0.0352 -0.0004 0. 0.2695 4.0711 0.8829 0.1175 6.8789 0. -0.2688 -0.008251 -0.001812 5.1954 -0.8501 -0.2405 -0.006704 0.207277 0.001 300 1.3514 0.6362 -0.8999 0.408 800. 0.4363 0.3119 0.5395 0.4338 0.7997 0.5307 0.4 -1.3794 0.165 -0.2146 -0.165 0.2146 6.5459 5.2317 1.06 2.4709 -2.1 -0.5 50 7.5735 3.6945 0.4807 0.0352 -0.0155 0. 0.2695 4.3745 0.8302 0.106 6.5334 0. -0.1793 -0.007267 -0.001274 5.0899 -0.9431 -0.1975 -0.007125 0.165464 0.004 300 1.4051 0.6049 -0.9618 0.462 800. 0.4396 0.3165 0.5422 0.4399 0.7988 0.531 0.5 -1.6508 0.165 -0.1972 -0.165 0.1972 5.2305 5.2893 1.06 2.5567 -2.1 -0.5 50 7.5778 3.7401 0.4707 0.0352 -0.0278 0.0991 0.2695 4.6099 0.7884 0.1061 6.526 0. -0.1428 -0.006492 -0.001074 4.7854 -1.0044 -0.1633 -0.007435 0.133828 0.01 300 1.4402 0.5507 -0.9945 0.524 800. 0.4419 0.3199 0.5433 0.4446 0.7966 0.5313 0.75 -2.1511 0.165 -0.162 -0.165 0.162 3.7896 5.4109 1.06 2.6812 -2.1 -0.5 50 7.5808 3.7941 0.4575 0.0352 -0.0477 0.1982 0.2695 5.0376 0.6754 0.1 6.5 0. -0.1138 -0.005147 -0.001115 4.3304 -1.0602 -0.1028 -0.00812 0.085153 0.034 300 1.528 0.3582 -1.0225 0.658 800. 0.4459 0.3255 0.5294 0.4533 0.7792 0.5309 1 -2.5365 0.165 -0.14 -0.165 0.14 3.3024 5.5106 1.06 2.7474 -2.1 -0.5 50 7.5814 3.8144 0.4522 0.0352 -0.0559 0.2154 0.2695 5.3411 0.6196 0.1 6.5 0. -0.1062 -0.004277 -0.001197 4.1667 -1.0941 -0.0699 -0.008444 0.058595 0.067 300 1.6523 0.2003 -1.0002 0.78 800. 0.4484 0.3291 0.5105 0.4594 0.7504 0.5302 1.5 -3.0686 0.165 -0.1184 -0.165 0.1184 2.8498 5.6705 1.06 2.8161 -2.1 -0.5 50 7.5817 3.8284 0.4501 0.0352 -0.063 0.2154 0.2695 5.7688 0.5101 0.1 6.5 0. -0.102 -0.002979 -0.001675 4.0029 -1.1142 -0.0425 -0.007707 0.031787 0.143 300 1.8872 0.0356 -0.9245 0.96 800. 0.4515 0.3335 0.4783 0.468 0.7136 0.5276 2 -3.4148 0.1645 -0.11 -0.1645 0.11 2.5417 5.7981 1.06 2.8514 -2.1 -0.5 50 7.5818 3.833 0.45 0.0352 -0.0665 0.2154 0.2695 6.0723 0.3917 0.1 6.5 0. -0.1009 -0.002301 -0.002349 3.8949 -1.1154 -0.0302 -0.004792 0.019716 0.203 300 2.1348 0. -0.8626 1.11 800. 0.4534 0.3363 0.4681 0.4681 0.7035 0.5167 3 -3.9013 0.1168 -0.104 -0.1168 0.104 2.1488 5.9983 1.06 2.8875 -2.1 -0.5 50 7.5818 3.8361 0.45 0.016 -0.0516 0.2154 0.2695 6.5 0.1244 0.1 6.5 0. -0.1003 -0.001344 -0.003306 3.7928 -1.1081 -0.0129 -0.001828 0.009643 0.277 300 3.5752 0. -0.7882 1.291 800. 0.4558 0.3398 0.4617 0.4617 0.7006 0.4917 4 -4.2466 0.0732 -0.102 -0.0732 0.102 1.8957 6.1552 1.06 2.9058 -2.1 -0.5 50 7.5818 3.8369 0.45 0.0062 -0.0448 0.2154 0.2695 6.8035 0.0086 0.1 6.5 0. -0.1001 -0.001084 -0.003566 3.7443 -1.0603 -0.0016 -0.001523 0.005379 0.309 300 3.8646 0. -0.7195 1.387 800. 0.4574 0.3419 0.4571 0.4571 0.7001 0.4682 5 -4.5143 0.0484 -0.101 -0.0484 0.101 1.7228 6.2856 1.06 2.9169 -2.1 -0.5 50 7.5818 3.8376 0.45 0.0029 -0.0424 0.2154 0.2695 7.0389 0. 0.1 6.5 0. -0.1001 -0.00101 -0.00364 3.709 -0.9872 0. -0.00144 0.003223 0.321 300 3.7292 0. -0.656 1.433 800. 0.4584 0.3435 0.4535 0.4535 0.7 0.4517 7.5 -5.0009 0.022 -0.101 -0.022 0.101 1.5737 6.5428 1.06 2.932 -2.1 -0.5 50 7.5818 3.838 0.45 0.0007 -0.0348 0.2154 0.2695 7.4666 0. 0.1 6.5 0. -0.1 -0.000964 -0.003686 3.6632 -0.8274 0. -0.001369 0.001134 0.329 300 2.3763 0. -0.5202 1.46 800. 0.4601 0.3459 0.4471 0.4471 0.7 0.4167 10 -5.3461 0.0124 -0.1 -0.0124 0.1 1.5265 6.7415 1.06 2.9396 -2.1 -0.5 50 7.5818 3.838 0.45 0.0003 -0.0253 0.2154 0.2695 7.77 0. 0.1 6.5 0. -0.1 -0.00095 -0.0037 3.623 -0.7053 0. -0.001361 0.000515 0.33 300 1.7679 0. -0.4068 1.464 800. 0.4612 0.3474 0.4426 0.4426 0.7 0.3755 """)
[docs]class ChiouYoungs2014Japan(ChiouYoungs2014): """ Regionalisation of the Chiou & Youngs (2014) GMPE for use with the Japan far-field distance attuation scaling and site model """
[docs]class ChiouYoungs2014Italy(ChiouYoungs2014): """ Adaption of the Chiou & Youngs (2014) GMPE for the the Italy far-field attenuation scaling, but assuming the California site amplification model """
[docs]class ChiouYoungs2014Wenchuan(ChiouYoungs2014): """ Adaption of the Chiou & Youngs (2014) GMPE for the Wenchuan far-field attenuation scaling, but assuming the California site amplification model. It should be note that according to Chiou & Youngs (2014) this adjustment is calibrated only for the M7.9 Wenchuan earthquake, so application to other scenarios is at the user's own risk """
[docs]class ChiouYoungs2014PEER(ChiouYoungs2014): """ This implements the Chiou & Youngs (2014) GMPE for use with the PEER tests. In this version the total standard deviation is fixed at 0.65 """ #: Only the total standars deviation is defined DEFINED_FOR_STANDARD_DEVIATION_TYPES = {const.StdDev.TOTAL} #: The PEER tests requires only PGA DEFINED_FOR_INTENSITY_MEASURE_TYPES = {PGA}
[docs]class ChiouYoungs2014NearFaultEffect(ChiouYoungs2014): """ This implements the Chiou & Youngs (2014) GMPE include the near fault effect prediction. In this version, we add the distance measure, rcdpp for directivity prediction. """ #: Required distance measures are RRup, Rjb, Rx, and Rcdpp REQUIRES_DISTANCES = {'rrup', 'rjb', 'rx', 'rcdpp'}
[docs]class ChiouYoungs2014ACME2019(ChiouYoungs2014): """ Implements a modified version of the CY2014 GMM. Main changes: - Hanging wall term excluded - Centered Ztor = 0 - Centered Dpp = 0 """ adapted = True
[docs] def compute(self, ctx: np.recarray, imts, mean, sig, tau, phi): """ See :meth:`superclass method <.base.GroundShakingIntensityModel.compute>` for spec of input and result values. """ for m, imt in enumerate(imts): C = self.COEFFS[imt] # intensity on a reference soil is used for both mean # and stddev calculations. ln_y_ref = _get_ln_y_ref(self.__class__.__name__, ctx, C) # exp1 and exp2 are parts of eq. 12 and eq. 13, # calculate it once for both. exp1 = np.exp(C['phi3'] * (ctx.vs30.clip(-np.inf, 1130) - 360)) exp2 = np.exp(C['phi3'] * (1130 - 360)) mean[m] = _get_mean(ctx, C, ln_y_ref, exp1, exp2)